1. Natural Language Processing DCG and Syntax NLP DCG A “translation” example: special case A DCG recogniser

2. Natural Language Processing NLP is the art and science of getting computers to understand natural language. NLP draws on materials from other disciplines: computer science, formal philosophy and formal linguistics NLP is an “AI complete” task: all the activities which turn up elsewhere in AI, such as knowledge representation, planning, inference and so on turn up in one form or another in NLP.

3. DCG (Definite Clause Grammar) An example $120pw $200perweek $150pweek [$, 1, 2, 0, p, w] price-->dollar, number, unit. dollar-->[$]. number-->digit, number. number-->digit. unit-->[p,w]. unit-->[p, e, r, w, e, e, k]. unit-->[p, w, e, e, k]. digit-->[1]. digit-->[2]. digit-->[3].... ?- price([$, 2, 3, 4, p, w], []). Yes ?-price([8, 0, 0, p, w, e, e, k], []). No.

4. Expand DCG to standard predicates Price-->dollar, number, unit. price(List1, List2):- dollar(List1, List11), number(List11, List12), unit(List12, List2). dollar-->[$]. dollar([$|List], List). digit-->[1]. digit([1|List], List]). number-->digit, number. number-->digit.

5. Extending DCG 1.Add variables 2.Add normal predicates in { } ?-price(X, [$, 1,2,3, p, w], []). X=[1,2,3] price(X)-->dollar, number(X), unit. number([D|T])-->digit(D), number(T). number([D])-->digit(D). digit(1)-->[1]. digit(2)-->[2]. price(X)-->dollar, number(X), unit, {length(X,N), N<3}.

6. Expand extended DCG to standard predicates price(X)-->dollar, number(X), unit, {length(X, N), N<3}. price(X, List1, List2):- dollar(List1, List11), number(X, List11, List12), unit(List12, List2), length(X, N), N<3.

7. A “machine translation” example three hundred and thirty four: 334 twenty one: 21 fourteen: 14 five: 5 ?-to_number(N, [three, hundred, and, thirty, four],[]). N=334.

8. A “translation” example Vocabulary, lexicon digit(1) --> [one]. digit(2) --> [two]. ….. digit(9) --> [nine]. teen(10) --> [ten]. teen(11) --> [eleven]. ….. teen(19) --> [nineteen]. tens(20) --> [twenty]. tens(30) --> [thirty]. ….. tens(90) --> [ninety].

9. A “translation” example Numbers with one or two digits. xx(N) --> digit(N). xx(N) --> teen(N). xx(N) --> tens(T), rest_xx(N1), {N is T+N1}. rest_xx(N) --> digit(N). rest_xx(0) --> [].

10. A “translation” example % numbers with 3 or fewer digits xxx(N) --> digit(D), [hundred], rest_xxx(N1), {N is D*100+N1}. xxx(N) --> xx(N). rest_xxx(N) --> [and], xx(N). rest_xxx(0) --> []. %top level to_number(0) --> [zero]. to_number(N) --> xxx(N). Query ?-to_number(N, [two, hundred, and, twenty one], []). N=221

11. Representing Syntactic Knowledge Syntactic knowledge: – Syntactic Categories: e.g. Noun, Sentence. – Grammatical features: e.g. Singular, Plural – Grammar rules. Why bother?

12. Parts of language Regard sentences as being built out of constituents Two types of constituents: – words (simple constituents), which have lexical categories like noun, verb, etc. – phrases (compound constituents), like noun phrases, verb phrases, etc. How to store syntactic knowledge? – lexicon – grammar rules

13. Words: Lexical Categories (Parts of Speech) Noun (N): Jack, tree, house, cannon Verb (V): build, walk, kill Adjective (Adj): big, red, unpleasant Determiner (Det): the, a, which, that – Jack built {the, a, that} big, red house; – Which house did Jack build? Preposition (Prep): with, for, in, from, to, through, via, under

14. Words: Lexical Categories (ctd) Pronoun (Pro): her, him, she, itself, that, it – I saw the man in the park with the telescope – Don't do that to him Conjunction (Conj): and, or, but. Two kinds of lexical categories: 1.Open categories (“content words”): N, V, Adj 2.Closed categories (“function words”): Det, Prep, Pro, Conj

15. Compound Constituents Some compound constituents: Sentence (S): Jack built the house. Noun Phrase (NP): John; the big, red house; the house that Jack built; the destruction of the city. Verb Phrase (VP): built the house quickly; saw the man in the park. Prepositional Phrase (PP): with the telescope; on the table

16. A Simple Grammar S  NP VP VP  V NP NP  Proper_N NP  det N Proper_N  John Proper_N  Mary N  cake V  loves V  ate det  the Sentences in this language: “John loves Mary” “John ate the cake” “John loves the cake”

17. Definite Clause Grammars (DCGs) The above grammar can be simply implemented in DCG notation as follows: s --> np, vp. vp --> v, np. np --> proper_n. np --> det, n. proper_n --> [john]. proper_n --> [mary]. n --> [cake]. v --> [loves]. v --> [ate]. det --> [the].

18. A Second Implementation A more efficient implementation: s --> np, vp. vp --> [V],{verb(V)},np. np --> [N],{proper_n(N)}. np --> [Det],{det(Det)}, [N],{noun(N)}. proper_n(john). proper_n(mary). noun(cake). verb(loves). verb(ate). det(the). Notes: 1.The {} allow Prolog code to be embedded in DCG rules 2.The cost of processing is now less dependent on lexicon size (given good indexing) 3.So far we have implemented only a DCG recogniser, not a parser.

19. Translating DCG Consider the rule s --> np, vp. Prolog translates this as: s(Ws1,Ws2) :- np(Ws1,Ws),vp(Ws,Ws2). This says that after taking an s off the start of Ws1, Ws2 remains The rule proper_n --> [john]. is translated as proper_n([john|Ws],Ws). Query s([john, ate, the cake],[]). Yes s([ate, john, cake, the],[]). No

20. Parsing How to represent a tree in Prolog? s npvp proper_nvnp detn [the][cake] [ate][john] Consider the goal s([john,ate,the,cake],[]) It will have the following parse/proof tree

21. A DCG Parser We can include extra arguments on non- terminals in our grammar, these allow us to record the parse tree: s(s(NP,VP)) --> np(NP), vp(VP). vp(vp(verb(V), NP)) --> [V], {verb(V)}, np(NP). np(np(proper_n(N))) --> [N], {proper_n(N)}. np(np(det(Det), noun(N))) --> [Det], {det(Det)}, [N], noun(N).

22. A DCG Parser (ctd.) Arguments to DCG non-terminals are expanded by Prolog like this: s(s(NP,VP),Ws1,Ws2) :- np(NP,Ws1,Ws),vp(VP,Ws,Ws2). Here, arguments are used to build up a parse tree: ?- s(Parse, [john,ate,the,cake], []). Parse = s(np(proper_n(john)), vp(verb(ate), np(det(the), noun(cake))))

23. Eliminating Left Recursion Example with left recursion: S  S, PP S  NP, VP Will get into infinite loop! Example with left recursion eliminated: S  S1, S-PP S-PP   S-PP  PP, S-PP S1  NP, VP (The difference is even clearer with 2 PPs – try drawing the 2 trees for NP VP PP PP) Here's how to get the same parse tree as in the 1st grammar, but using the 2nd grammar: s(S) --> s1(S1), s_pp(S1,S). s_pp(S,S) --> []. s_pp(S0,S) --> pp(PP), s_pp(s(S0,PP),S). s1(s(NP,VP)) --> np(NP), vp(VP). S S1 NP VP S-PP PP S-PP  S S NP VP PP

24. Problems of the simple parser